He turned approver. The judge still didn't trust him alone.

The accomplice confessed. The court believed him. But the judge said: not enough.

Sarwan Singh had turned approver — a co-accused who agreed to testify for the prosecution in exchange for a pardon. He stood in the dock alongside Gurdayal Singh and Harbans Singh, all charged with murder. The pardon document, signed in ink, promised him freedom if he spoke. And he did speak — of the plan, the murder, the roles each man played. The trial court found his account credible. Yet when the time came to convict the remaining accused, the court paused. One man's word, even a confessed participant's, could not be the sole foundation for a murder conviction. The Supreme Court would soon agree — and in doing so, lay down a rule that still governs every accomplice testimony in India today.

The question was simple but brutal: Can a man be sent to prison — or to the gallows — based entirely on the word of someone who admits he was part of the crime?

Banta Singh took the stand

In Sarwan Singh v. State of Punjab, the Supreme Court was dealing with a joint trial for murder. Several accused stood together in the dock: Sarwan Singh, Gurdayal Singh, and Harbans Singh. The prosecution's case rested heavily on the testimony of Banta Singh — a man who was himself implicated in the crime but had been granted a pardon (a full release from prosecution) to testify against his former co-accused. The courtroom fell silent as Banta Singh took the stand. The transcript of his testimony, stacked on the clerk's table, carried the weight of the case. He described the murder in detail. He named names. He gave times and places. The trial court believed him. All the accused were convicted. But when the case reached the Supreme Court, the bench — the panel of judges hearing the appeal — took a harder look at what it meant to convict a person on an accomplice's word alone.

The problem with a pardoned witness

An accomplice who turns approver has every reason to lie. He has been promised freedom in exchange for testimony. The more convincingly he implicates others, the more valuable his evidence becomes to the prosecution. A judge who accepts such testimony without question risks sending an innocent man to prison while the real culprit walks free — rewarded for his cooperation.

The Supreme Court recognised this danger. It held that while a conviction based solely on an accomplice's uncorroborated testimony is not technically illegal — the law does not forbid it — it is, as a matter of prudence, unsafe. The Court put it plainly: judges must exercise "extreme caution and care" when considering accomplice evidence. What began as a rule of practice, the Court said, had acquired the force of a rule of law. The bench looked at Banta Singh's testimony and asked: what else is there?

The first test: Is the accomplice believable?

The Court laid down what has become known as the "double test" for approver evidence. The first test applies to every witness in every trial: reliability. The accomplice's story must be examined for inherent improbabilities. Does it make sense? Does it contradict itself? Is there any reason to believe the witness is lying?

The Court held that the evidence of the accomplice should contain "nothing inherent or improbable." The High Court, in reviewing the trial court's decision, must find no basis for concluding that the approver gave false evidence. This is the baseline — the minimum threshold any witness must cross before their testimony can be acted upon. In Sarwan Singh, the Court stressed that the evidence of an accomplice is always subject to this scrutiny. The Court examined Banta Singh's testimony for inherent improbabilities, as required by the first test — the sequence of events, the specific roles — and asked whether a man who had confessed to his own part could be trusted to tell the whole truth about the others.

The second test: Where is the corroboration?

But for an accomplice, reliability alone is not enough. The Court imposed a second requirement: even if the witness is deemed reliable, "his evidence must receive sufficient corroboration." This corroboration must go to "material particulars" — the key facts that establish the accused's involvement in the crime.

Corroboration does not mean the prosecution must produce a second eyewitness to every detail. It means there must be independent evidence — a piece of physical evidence, a credible witness who is not an accomplice, a document, a circumstance — that supports the accomplice's account on the essential points. Without it, the conviction is unsafe. The Supreme Court bench, after finding Banta Singh reliable, then turned to the rest of the record. It scanned the witness list, the exhibits, the post-mortem report. Where was the corroboration? The third accused, Sarwan Singh, had become an approver later, but his testimony was not the foundation. That foundation was Banta Singh's word alone — and the Court wanted more.

Why the double test matters

The double test protects against a specific kind of injustice: the conviction of an innocent person based on the word of a motivated liar. An accomplice who has been pardoned has a powerful incentive to shift blame, exaggerate, or fabricate. The double test forces the court to look beyond the accomplice's narrative and ask: What else is there?

In Lachi Ram v. State of Punjab, a later case that applied the same principle, the Supreme Court examined whether the double test had been properly satisfied. The accused had been convicted of murder based on an approver's evidence that was corroborated on material particulars. The Court found that the approver's testimony was indeed reliable — the first test was met — and that good prosecution witnesses had provided the necessary corroboration. The conviction stood. The Court had observed the logic from Sarwan Singh: "an approver's evidence to be accepted must satisfy two tests."

But the principle remains: one without the other is not enough. Reliability without corroboration leaves the door open to a single motivated witness sending a man to prison. Corroboration without reliability — evidence that supports a story the witness never told — is equally meaningless.

The double test is not a mere procedural formality. It is a safeguard that has been refined through decades of judicial experience. In Sarwan Singh, the Supreme Court did not merely announce a rule — it traced the reasoning behind it. The Court noted that the evidence of an accomplice is always subject to corroboration, not because the law commands it, but because prudence demands it. Over time, that rule of practice hardened into a rule of law, binding on every trial court and High Court in the country.

Consider the procedural context. A joint trial involves multiple accused, each with their own defence. The prosecution must prove its case beyond a reasonable doubt against each one. When an approver takes the stand, the defence has the right to cross-examine — to test the witness's credibility, to expose inconsistencies, to challenge motives. But cross-examination alone may not be enough. The double test provides an additional layer of protection: even if the approver survives cross-examination, the court must still look for independent corroboration.

The Lachi Ram case illustrates how the double test works in practice. There, the Supreme Court examined whether the approver's evidence had been corroborated on material particulars. The Court found that the approver was reliable — his account contained nothing inherently improbable. But the Court did not stop there. It then looked to the prosecution witnesses who had provided independent evidence supporting the approver's story. Because both tests were satisfied, the conviction was upheld.

What happens when the double test is not met? The conviction falls. The accused walks free. That is the consequence of a rule designed to prevent miscarriages of justice. The Supreme Court has made clear that a conviction based on uncorroborated accomplice evidence, while not illegal, is unsafe. No judge should risk sending a man to prison — or to the gallows — on such fragile ground.

The accomplice's testimony, however compelling, remains the word of a man who has already admitted his own guilt. He has been promised a pardon. He has every reason to please the prosecution. The double test ensures that the court does not become a tool for that motivation.

The accomplice confessed. The court believed him. But the judge said: not enough — and the law agreed.

THE TEST: Before a court convicts on an accomplice's testimony, it must ask two questions — first, is this witness reliable, and second, is there independent evidence supporting the key facts of the crime?